Efficient feature-based image registration by mapping sparsified surfaces

摘要

With the advancement in the digital camera technology, the use of highresolution images and videos has been widespread in the modern society. Inparticular, image and video frame registration is frequently applied incomputer graphics and film production. However, the conventional registrationapproaches usually require long computational time for high quality images andvideo frames. This hinders the applications of the registration approaches inthe modern industries. In this work, we propose a novel approach called {emTRIM} to accelerate the computations of the registration by triangulating theimages. More specifically, given a high resolution image or video frame, wecompute an optimal coarse triangulation which captures the important featuresof the image. Then, the computation of the registration can be simplified withthe aid of the coarse triangulation. Experimental results suggest that thecomputational time of the registration is significantly reduced using ourtriangulation-based approach, meanwhile the accuracy of the registration iswell retained when compared with the conventional grid-based approach.